NIPL: Nemo's Idea for a Play List
This playlist is based on several assumptions about how humans listen to music and organise and generate their own playlists. Please read the NIPL/Assumptions page before continuing.
So the NIPL system, broadly, is:
- Users rate not just the song, but the song->nextsong pair. (eg, Rammstein and Enya might both get high scores on their own, but Enya->Rammstein is likely to generate a low score)
- With advanced use of MarkovChains and other sorting cleverness, playlists are generated based on song-pair scores, not just song scores)
Some component parts of NIPL include:
- Genres (ideas for a comprehensive genre classification system)
- NUDI (Nemo's Ultimate Disc Identifier)
- WeightedRandom (The ability to choose the next song in a "random" but usefull manner)
There are a number of IRC logs (well 4) from August and September 2001. In these, NIPL ideas are brainstormed and thrashed out in some detail. To really get a good idea of NIPL, it's worth reading through those. Especially as it's quite likely we've covered alot of potential questions ourselves already.
Logs can be found at http://pub.thorx.net/nipl/irclogs/
There is an NIPL/Implementation page with old notes of database structure and the like. In 2002 there was a an implementation of NIPL for Windows, named Project Playlist. Beta2 was released to public, and can be seen on archive.org (I've not checked if the .zip is there)
- Julian Orback on facebook - my friend Julian came up with the same basic idea "choosing which of the songs you like to play next" and posted his thoughts, including finding the following links - the first two described as "near miss" to his idea, and the third a direct hit.
- https://github.com/BigFav/Markov-Music-Shuffler (trains on genre/song similarity)
- http://www.mirlab.org/conference_papers/International_Conference/ISMIR%202011/papers/PS4-11.pdf (judges an existing playlists)
- https://www.cs.cornell.edu/people/tj/publications/chen_etal_12a.pdf (trains on existing playlists)
- "Boil the Frog" uses spotify "similar artist" measure to create smooth transitions. It's a completely different method to the NIPL proposal, but the end result is pretty similar!
Older links (pre-2009, not checked since):
- Synapse - A similar idea?
- GJay - Other ways of generating a good next-song method.
- iRate - a ratings songs, not links, but collaboratively
- Building a Music Map - some good discussions here - including from Nemo
NIPL is in accordance with NINS